Climate models project large changes in rainfall, but disagree on their magnitude and sign. The consequences of this uncertainty on optimal dam dimensioning is assessed for a small mountainous catchment in Greece. Optimal dam design is estimated using a Cost-Benefit Analysis (CBA) based on trends in seasonal temperature and precipitations from 19 IPCC-AR4 climate models driven by the the SRES A2 emission scenario. Optimal reservoir volumes are modified by climate change, leading to up to 34% differences between optimal volumes. Contrary to widely-used target-based approaches, the CBA suggests that reduced rainfall should lead to smaller water reservoirs. The resulting change in the Net Present Value (NPV) of water supply is also substantial, ranging from no change to a large 25% loss, depending on the climate model, even assuming optimal adaptation and perfect foresight. In addition, climate change uncertainty can lead to design errors, with a cost ranging from 0.3 to 2.8% of the NPV, depending on site characteristics. This paper proposes to complement the CBA with a robust decision-making approach that focuses on reducing design-error costs. It also suggests that climate change impacts in the water sector may reveal large, that water reservoirs do not always provide a cost-efficient adaptation strategy, and that alternative adaptation strategies based on water conservation and non-conventional water production need to be considered.